II. The appeal and underlying problem of self-exclusion

Amongst Johnny Cash’s voluminous catalogue is a song entitled When Uncle Bill Quit Dope.Footnote ¹⁹ Though more obscure than some of his vaunted classics, Cash presents in this song a story of remarkable emotional and, indeed, philosophical depth. ^{Footnote 20} He tells the story of a cocaine-addicted man who implores his sister, Louise, to lock him in an upstairs room and refuse to release him until “[he] don’t want dope no more.”Footnote ²¹ After fourteen arduous days and nights, Louise opens the door of Bill’s makeshift prison cell to find him in a renewed state of health and happiness, having overcome his addiction to “that demon cocaine.”Footnote ²² The covenant Cash describes between Uncle Bill and Louise is, in fact, a Ulysses pact, named for the legendary Greek hero known in Greek as Odysseus. Ulysses pacts derive their name from a particular incident in Homer’s Odyssey. Ulysses, keen to hear the Sirens’ song but aware of the ruination that might accompany it, fills the ears of the sailors under his command with wax, temporarily deafening them, and instructs them to tie him to the ship’s mast and not release him even if he later orders them to. Upon hearing the Sirens’ enticements, Ulysses is overcome with temptation and protests to his men to release him but, relying on his prior instruction, they do not. ^{Footnote 23} The comparison with self-exclusion agreements is obvious. In the same manner as Uncle Bill and Ulysses, the gambler acknowledges that they will face, at some future time, an urge to engage in behaviour they presently think imprudent. In order to forestall this future behaviour, they decide in advance to enlist others to prevent their subsequent indulgence. In other words, the gambler decides at one time (t ¹ ) that they should be prevented from gambling at a later time (t ² ) and requests (or, in jurisdictions where providing self-exclusion is mandatory, demands) that others (gambling operators and their staff) do so.

Johnny Cash is not alone in recognising the applicability of Ulysses pacts to addiction. ^{Footnote 24} In Ulysses and the Sirens, for example, Jon Elster argues that humans exhibit an imperfect rationality. ^{Footnote 25} Ulysses rationally knows that he should not succumb to the Sirens’ enticements and would, if he were a perfectly rational actor, need no more than that knowledge to resist them. His capacity to act on this knowledge, however, is impaired by non-rational factors in his human constitution, such as the weakness of his will and overwhelming emotion. Elster elsewhere opines that addictive behaviour is “the most striking instance … of ambivalence or weakness of will.” ^{Footnote 26} Because the addict often experiences the pattern of attempted recovery and relapse multiple times, they realise that they must “confront not only their addiction, but their inability to quit by simply deciding to do so.” ^{Footnote 27} Precommitment strategies, including Ulysses pacts, are a logical solution to this predicament. Inasmuch as addiction is caused by rationality-disrupting emotional factors, it is part of a strand of thought almost as old as Ulysses’ own pact with his sailors—akrasia. In Nicomachean Ethics, Aristotle observes that an agent may act irrationally because they “have the rational principle but do not abide by it, since [they] are defeated by a weaker passion, and do not act without previous deliberation” ^{Footnote 28} or “by reason of their quickness [or] the violence of their passions do not await the argument, because they are apt to follow their imagination.” ^{Footnote 29} This notion of an akratic (i.e. weak-willed or emotional and, ultimately, irrational) agent is useful in explaining addicts’ relapse and understanding the value of Ulysses pacts in addressing it. Heather and Segal, echoing Elster, consider addiction to be a “paradigm case for the study of weakness of will.” ^{Footnote 30} They argue that the notion of an akratic addict refutes what they call the strong sense of the causal attribution hypothesis of addiction. ^{Footnote 31} Advanced inter alia by Davies, this hypothesis contends that addiction is a myth and that the behaviours of putative addicts are as voluntary as any other personal choice. ^{Footnote 32} Drawing on the work of Donald Davidson, ^{Footnote 33} Heather and Segal explain the possible irrationality of an addict’s relapse with the example of a smoker who is trying to quit. The smoker has, as Davies contends, a reason (r) to relapse, namely that it will bring immediate enjoyment. Knowing that immediate enjoyment is not the only relevant interest, however, the smoker has a more comprehensive reason (r’) not to relapse. r’, being a reason based on all the information presently available to the smoker, includes r. If the smoker relapses, this behaviour is akratic and irrational because it is motivated by an “insufficient” reason, r, in comparison to the overall better reason, r’, which subsumes r. ^{Footnote 34} This outline of akrasia is easily reconcilable with the behaviours of gambling addicts. Repeated failed attempts to abstain from gambling are a diagnostic criterion for Gambling Disorder (GD). ^{Footnote 35} More obviously, gambling addicts suffer from a range of rationality-diminishing factors, such as cravings, trance-like states, and even memory blackouts. ^{Footnote 36} Gambling Disorder is also strongly related to a high degree of impulsivity. ^{Footnote 37} All this suggests that gambling addicts are likely to choose r, the inferior reason to gamble, over r’, the superior, all-things-considered reason to abstain.

In view of what I have outlined thus far, self-exclusion has obvious value as a mechanism to aid gambling addicts. However, many unresolved issues remain, and it is these issues which will form the remainder of this portion of the present article. Gambling self-exclusion differs in key ways from the Ulysses pacts successfully deployed by Uncle Bill and by Ulysses himself. The ad hoc, informal arrangement between Uncle Bill and Louise is distinguishable from the formal, legal arrangement self-exclusion represents. Indeed, Louise, despite her good intentions, is probably guilty of false imprisonment. ^{Footnote 38} Gambling self-exclusion differs in an even starker and, for the purposes of this article, more important way from Ulysses’ agreement with his sailors. Because of Ulysses’ superior rank, paternalism is not a possible alternative to his Ulysses pact. Any sailor or group of sailors who tied Ulysses to the mast without his consent, whether present or prior, would be guilty of insubordination. This is not the case with a problem gambler who seeks to self-exclude. Paternalism, in the form of forced exclusion, is a possible, as distinct from being a desirable, option for policy-makers. If self-exclusion is to have a unique merit, it is, as I earlier posited, found in its basis in the self-excluder’s autonomy. ^{Footnote 39} While I will discuss autonomy in greater depth later, grounding self-exclusion, or any measure, in autonomy allows it to avoid the issues inherent in paternalism, in particular its condescension and the affirmative value judgement upon which it relies.

Criticisms that paternalism is condescending are well-founded and many, though they do not warrant full attention here. For Anderson, some forms of paternalism are tantamount to “telling citizens that they are too stupid to run their lives.” ^{Footnote 40} For Feinberg, the “arrogant and demeaning” practice of paternalism, whereby adults are treated as children by a state standing in loco parentis, threatens to deprive citizens of their decision-making power and infantilise them. ^{Footnote 41} Self-exclusion purports to avoid this issue. To be sure, a self-excluded gambler being refused service appears outwardly indistinguishable to one being paternalistically refused service. The former, however, avoids the condescension of the latter because the refusal is not justified, to use Feinberg’s terminology, because “Daddy knows best” ^{Footnote 42} but because the gambler knew best when they executed the self-exclusion agreement.

The second criticism of paternalism which self-exclusion purports to avoid is that it relies on an indefensible value-judgement. Forced exclusion, for example, relies on the paternalist’s affirmative judgement that the excluded party should not gamble. The rightness of this judgement, critics argue, is indemonstrable and, ultimately, beyond the paternalist’s limited knowledge. As Mill famously wrote, “[s]peaking generally, there is no one so fit to conduct any business, or to determine how or by whom it shall be conducted, as those who are personally interested in it.” ^{Footnote 43} Mill argues, in essence, that the prospective paternalised person is always a better judge of what is right for them than the paternalist. This is a highly suspect assumption. The above discussion of akrasia highlights one way in which a problem gambler is, at least, not a very good judge of the right course of action. The akratic gambler is driven by what Loewenstein calls ‘visceral’ influences, which impede their ability to make a rational decision. ^{Footnote 44} In addition to visceral, emotional influences on decision-making, a wealth of research in behavioural economics has also demonstrated the pervasive role of cognitive biases in impeding rational decision-making. A previously dominant assumption in economic theory was the existence of a homo economicus, an agent who acts perfectly rationally and self-interestedly, based on complete information. More recently, however, a wealth of research in behavioural economics has severely undermined this assumption. ^{Footnote 45} Rather, the rationality of decision-making can be severely disrupted by various cognitive biases, including status quo bias, the framing effect, and the use of representative heuristics. ^{Footnote 46} This, again, has significant application to problem gamblers. Inasmuch as problem gamblers are susceptible to akratic, visceral urges, they also exhibit a marked susceptibility to cognitive bias. ^{Footnote 47} Although the ‘best judge’ argument against paternalism is undermined by the presence of rationality-subverting visceral urges and cognitive biases, this does not necessarily justify paternalism. Those in positions of authority, including would-be paternalists, are also imperfect judges. Research has demonstrated that civil servants, asylum judges, emergency managers, and holders of other similar positions are also susceptible to a range of cognitive biases. ^{Footnote 48} Though Bastiat did not have access to this research, he may as well have when he wrote, “[s]ince the natural tendencies of mankind are so bad that it is not safe to allow them liberty, how comes it to pass that the tendencies of organizers are always good?” ^{Footnote 49} The answer, of course, is that the tendencies of paternalists will not always be good or, at least, grounded in an accurate appreciation of the facts. The limitations of prospective paternalists surpass mere cognitive bias, however. Rather, the paternalist’s legitimacy is undermined by their lack of absolute knowledge of objectively good ends or, at least, the indemonstrability of those ends. To be sure, the relative susceptibility of paternalists and problem gamblers to the framing effect can be ascertained and compared, and the better judge discovered. Likewise, the accuracy of their respective knowledge of the odds of roulette can be demonstrated. The paternalist’s claim to knowledge of the objective rightness or wrongness of gambling (whether generally or in the case of a particular gambler) cannot be demonstrated in a way which would satisfy the evidentiary criteria of the liberal state. ^{Footnote 50} The indemonstrability of objectively good ends is why Le Grand and New reject what they call ends-related or moral paternalism. ^{Footnote 51} Because objectively good ends cannot be supported by empirical evidence, ends-related paternalism will invariably rely on the value judgements of the paternalist.

If self-exclusion is to be prized for its basis in the gambler’s autonomy, it must avoid these objections. The objection from condescension is presumably mitigated to some degree by the problem gambler’s prior indication at t ¹ that they ought not gamble at t ² . Those enlisted at t ² to prevent the self-excluder from gambling are not telling the self-excluder that “they are too stupid to run their [life]” but are instead saying that the self-excluder is wise enough to run their life and did so at t ¹ . ^{Footnote 52} On the other hand, the objection from value-partiality—and with it the self-exclusion problem—remain. Those who would prevent a t ² self-excluder from gambling are faced with the choice of whether or not to refuse service. ^{Footnote 53} If self-exclusion is to regard the self-excluder’s autonomy, those enforcing it must act on the self-excluder’s preferences and not their own. However, the self-excluder has two conflicting preferences. The t ¹ preference is for the gambler to be refused service at t ² , while the t ² preference is for the gambler to be allowed to gamble at t ² . If self-exclusion is to exist at all, the t ¹ preference must, of course, take precedence and the t ² preference must be discarded. How can this precedence be justified, however? It cannot, for the reasons I have outlined, be because the t ¹ preference is ‘correct.’ Another justification might be an arbitrary favour for earlier decisions over later ones. This is also unfeasible because, as much as it would enable the operation of self-exclusion, it would equally justify the enforcement of agreements by gamblers never to self-exclude, perhaps included in the terms and conditions of gambling contracts. An ostensibly more meritorious justification is to find some way in which the t ² decision fails to satisfy the requirements of an autonomous decision. In other words, if the t ² decision-maker is not fully capable to decide to gamble, the t ² decision can be discarded and the t ¹ decision acted on.

Capacity is, after all, the “gatekeeper of autonomy,” ^{Footnote 54} and even Mill, the arch-libertarian of the nineteenth century, withheld autonomy from those not “in the maturity of their faculties” and those who are “delirious, or in some state of excitement or absorption incompatible with the full use of the reflecting faculty.” ^{Footnote 55} The visceral urges which bear on the akratic problem gambler could easily be seen as making their decision to gamble less-than-rational, less-than-autonomous, or less-than-capable. Nonetheless, the concept of legal capacity is fraught with complexity and is ever-changing. ^{Footnote 56} If capacity is to be equated, rudimentarily, with “understanding,” ^{Footnote 57} then its applicability to Gambling Disorder is contestable. GD sufferers are not tout court incapable in a general sense. A recent study, for example, found that GD sufferers are highly susceptible to the effect of positive framing in the context of financial decision-making. ^{Footnote 58} With regard to healthcare-related decisions, however, sufferers were no more susceptible to the framing effect than the control group. On the other hand, a functional approach to capacity, as adopted in the Mental Capacity Act 2005 in England and Wales, does not demand general incapacity. ^{Footnote 59} Rather, a person may lack capacity in certain spheres and enjoy it in others. A GD sufferer may therefore lack capacity to enter gambling contracts but have full capacity to enter a mortgage contract or choose whether or not to undergo medical treatment. The question of whether a problem gambler has capacity, however, is only relevant to justifying forced exclusion. In the realm of self-exclusion, it is beside the point. If the t ² decision is made by an incapable agent and therefore non-autonomous, then the t ¹ decision is irrelevant and the incapable gambler ought to be prevented from gambling whether they expressed a contrary prior preference or not. If, as I have already said, the allure of self-exclusion is its basis in autonomy, the t ¹ preference is crucial and cannot be rendered irrelevant to the refusal of service at t ² .

Before proceeding to the next portion of this article, it is worth restating what has been discussed thus far. Self-exclusion, as a Ulysses pact, has obvious appeal in addressing problem gambling. The problem gambler cannot, in many cases, abstain from gambling by sheer force of will. Rather, rationality-impugning forces may cause the gambler’s commitment to their abstinence to waver. In anticipation of their future wavering, the gambler can enlist others to prevent them from reneging on their earlier commitment. If self-exclusion is meritorious in a way that is distinct from forced exclusion, however, it must be because it emphasises the self-excluder’s autonomy. If this is the case, the t ² refusal of service cannot be based on a value judgement of any party other than the self-excluder, nor can it be because the t ² decision to gamble is prima facie discardable. Simply put, the t ² refusal of service can only occur because of the t ¹ decision to self-exclude.

III. Self-exclusion and the inadequacy of procedural autonomy

A central assumption of this article is that self-exclusion is prized because it values the problem gambler’s autonomy, in contrast to paternalistic forced exclusion, which does not. The long and complex history of the term ‘autonomy’ is by now familiar to many and would be impossible to recount in full here. Etymologically, the term derives from the Ancient Greek auto, meaning self, and nomos, meaning law. To be autonomous in its simplest form, then, means to make one’s own law, to be governed by rules one has developed for oneself. As Gerald Dworkin notes, the term originally applied to Greek city-states that made their own laws free of external interference, before its “natural extension” to individuals as being autonomous when they are self-determining. ^{Footnote 60} Despite its common root, however, the term now encompasses a variety of different and often conflicting meanings. Returning to Gerald Dworkin, he writes that, “the only features held constant from one author to another are that autonomy is a feature of persons and that it is a desirable quality to have.” ^{Footnote 61} One useful way to divide the vast range of theories is into the categories of substantive, which place direct constraints on the content of decisions for them to count as autonomous, and procedural, which require only that decisions, whatever their content, are arrived at in a procedurally independent way. In this article, I take autonomy to refer to a procedural theory. First, I believe simply that this is what is instinctively understood as the meaning of autonomy by anyone not steeped in its voluminous theoretical treatment, including gambling operators and their staff, gambling regulators, legislators, and prospective self-excluders themselves. Self-exclusion is a fundamentally practical exercise involving two parties, the self-excluder and those who would enforce self-exclusion at t ² . As such, its theoretical underpinnings must be such that they are mutually understood by both parties. Second, a substantive theory of autonomy that considers the t ² decision to gamble as necessarily heteronomous nullifies the t ¹ decision in a way I have earlier rejected. ^{Footnote 62} If the t ² decision is declared heteronomous on a substantive account, the prospective gambler is excluded for reasons other than their t ¹ decision, which deprives self-exclusion of its apparently unique value. Finally, although this article is not intended as a refutation of substantive autonomy, questions linger about whether substantive theories can be properly categorised as autonomy at all. Specifically, the value judgements inherent in these theories may be entirely alien to the agent whose autonomy they purport to represent. In the extreme, we see theories such as Babbitt’s. ^{Footnote 63} She argues that the liberal view, which bases an agent’s autonomy on their “untransformed” self, is inadequate in cases where an agent does not have a self which would choose a “full sense of autonomy,” even in idealised conditions.Footnote ⁶⁴ Rather, Babbitt argues that the agent’s sense of self must be transformed into one which would endorse a full sense of autonomy, a contention with all the disconcerting implications of Rousseau’s plan for the citizens of his envisioned state. ^{Footnote 65} What Babbitt advocates cannot, of course, properly be called autonomy, as it proposes the destruction of the agent’s self (auto) and the imposition of another’s law (nomos) on the agent. Any theory of autonomy in this vein is, in fact, anathema to autonomy in its proper form and can be more accurately described as heteronomy or xenonomy. ^{Footnote 66}

Procedural autonomy is content-neutral. ^{Footnote 67} It does not place direct constraints on the substance or content of decisions, requiring only that an agent’s decisions “express her will, and they express her will because the dispositions which they put into effect are hers in some important sense.” ^{Footnote 68} The demand that the dispositions which inform a decision are the agent’s own gives effect to the most literal meaning of autonomy, that of self-imposed law. It disqualifies actions which are motivated by dispositions arising from others, a category which Gerald Dworkin takes to include hypnosis, manipulation, coercion, and subliminal influence. ^{Footnote 69} Insofar as procedural autonomy requires freedom from others’ influence, it echoes the negative liberty of Berlin’s dichotomy, which he defines simply as being free from interference by others with the range of activities that are within one’s capacity. ^{Footnote 70} Nonetheless, mere freedom from external interference fails to fully explain the demands of procedural autonomy. Indeed, this sort of de facto freedom would fail to preclude an agent’s life being a turbulent series of disjointed and whimsical decisions. ^{Footnote 71} In other words, it would focus narrowly on the auto and remain wholly ignorant of the nomos; it would promote an anarchy of the self rather than a governed self. As such, procedural conceptions of autonomy tend to demand not just freedom from interference but a degree of de facto self-government to which de jure liberty is blind. ^{Footnote 72} The agent is generally not considered to be autonomous simply because they act free from external constraints, but because they, in some sense, ‘own’ or ‘identify with’ their decision. Traces of this requirement are evident in Mill’s On Liberty, where he bemoans that individuals ask themselves, “What is usually done by persons of my station and pecuniary circumstances?” rather than “What would suit my character and disposition?” ^{Footnote 73} The question of determining whether a person’s decisions are truly reflective of their “character and disposition” has abided since Mill first identified it. One attempt to resolve this issue is with a structural, or hierarchical, approach to preferences. ^{Footnote 74} For Gerald Dworkin, autonomy involves not just first-order preferences but also “preferences about … preferences, a desire not to have or act upon various desires.” ^{Footnote 75} Autonomy, then, constitutes

The hierarchical theory of preferences Dworkin advances here is not the only such theory or even, as I have observed in the footnotes, the only such theory advanced by Dworkin himself. Nonetheless, the nuances which separate these various theories are largely irrelevant here and I provide Dworkin’s above theory for exemplary purposes only. What is relevant is the distinction between first- and second-order preferences and the requirement that the latter be regarded in some sense for an action to be autonomous. The structural approach appears, at first glance, to offer a solution to the self-exclusion problem. The t ¹ decision to self-exclude can be neatly equated with a second-order preference not to gamble and the t ² decision to gamble can then be categorised as a heteronomous first-order preference because it fails to cohere with or otherwise properly regard the superior second-order preference. ^{Footnote 77} This avoids the issue of value-partiality I objected to in other theories because both the first- and second-order preferences emanate from the agent and not another party.

Hierarchical theories of preferences are not without issue, however. First, as much as the validity of first-order preferences must be ascertained with reference to second-order preferences, the validity of the second-order preferences must also be determined. This can presumably only be accomplished with reference to third-order preferences, which must be measured against even higher-order preferences. This introduces the problem of “infinite regress.” ^{Footnote 78} Writing in a different but still relevant context, Elster provides the following example to question the reality of a simple two-preference model:

A simple two-preference model fails to explain the predicament of the prospective cake eater. The first-order preference is to eat the cake, which is heteronomous because of the second-order preference to remain slim. However, this is, in turn, heteronomous because of a third-order preference to reject vanity. Elster seems to suggest this third-order preference is unreliable because it is possibly situational and self-deceptive. (Although this may be framed as a fourth-order preference to make consistent, rational decisions.) Frankfurt, another proponent of a hierarchical model, acknowledges this potential shortcoming but proposes that a non-arbitrary tourniquet can be placed on the infinite regress. ^{Footnote 80} The infinite regress can, he says, be terminated when the agent identifies “decisively” with a first-order preference and this “commitment ‘resounds’ throughout the potentially endless array of higher orders.”Footnote ⁸¹ He provides the example of a person who “without reservation or conflict, wants to be motivated by the desire to concentrate on his work.”Footnote ⁸² Because his second-order volition to be so motivated is “decisive,” he can “properly insist that this question concerning a third-order desire does not arise.”Footnote ⁸³ Whether this is truly non-arbitrary, however, is deeply contestable. ^{Footnote 84} If Frankfurt is suggesting that the force of identification is sufficient to forestall subsequent orders of preference, it must be asked how force is to be measured. The force of a preference can only be mitigated by other preferences, not by factors internal to that preference. For example, two students may have respective second-order preferences to achieve good results. The force of these respective preferences must be assumed to be equal. If the force of one student’s preferences is sufficient to preclude higher-order preferences, it is because that student also has a first-order preference to study diligently. If the preferences of the other student are not sufficiently forceful, it is because that student has a first-order preference to procrastinate. Inevitable arbitrariness arises in determining how many orders of preference must be coherent before further investigation becomes unnecessary. Even if this is not what Frankfurt suggests, a greater problem remains. The fact that a commitment must resound “throughout the potentially endless array of higher orders” admits that these higher orders exist. The fact that an agent’s lower-order preference can resound so endlessly simply speaks to an agent whose preferences are so consistent that they would remain entirely monolithic through infinite orders. If agents that are non-fragmented in this way exist, they are surely in a minority. In any case, I do not believe that such an agent—invariably impervious to inertia, lethargy, distraction, etc.—exists.

Gerald Dworkin adopts a different tack to Frankfurt in dealing with the problem of infinite regress. ^{Footnote 85} He claims to analyse autonomous persons rather than autonomous acts. As such, a person is autonomous if they have the capacity for critical reflection and exercise it, so long as this exercise is itself free from third-party interference and the person possesses the requisite identification with their first-order preference. There is, Dworkin says, no “conceptual necessity” for questioning whether second-order preferences are themselves justified by still higher-order preferences.Footnote ⁸⁶ In fact, Dworkin abrogates the question. Of course, his theory for gauging the general autonomy of a person may have some broad value in determining if, for example, a person with anterograde amnesia is autonomous. ^{Footnote 87} The self-exclusion problem, however, contains far more nuance than simply determining whether a person is tout court autonomous. As discussed previously, there is reason to believe that the irrationality of problem gamblers is domain-specific. ^{Footnote 88} Adopting a Dworkinian approach to autonomy might categorise a problem gambler as an autonomous person but ignore that, with regard to gambling, they cannot or do not engage in critical reflection.

Even leaving aside these issues with the respective theories of Frankfurt and Dworkin, the hierarchical approach is beset by an even more fatal issue: its impracticability. As an abstract principle of philosophy, the hierarchical approach might be meritorious, notwithstanding the problem of infinite regress. As a practical solution to the self-exclusion problem, however, it is not. It might be tempting to assume that the t ¹ decision (or any other binding decision) is rational and/or representative of a higher-order preference and that the t ² decision (or any other bound decision) is irrational and/or representative of a mere first-order preference. This, however, is a third-party assumption which is not necessarily correct. A precommitting decision may, as Elster observes, be no more rational or dispassionate than the later decision the precommitment seeks to restrain. Elster gives the example of covenant marriage, a form of marriage which is more difficult to enter and dissolve than ordinary marriages. An engaged person might be motivated by a number of non-rational or first-order preferences to opt for a covenant marriage, such as intense infatuation or a reluctance to imply to their prospective spouse that their commitment is not absolute. ^{Footnote 89} In this instance, a first-order preference unsupported by second-order reflection might bind a later first-order preference (to divorce) which is supported by a second-order preference (self-respect or the value of marital faithfulness) when the person’s spouse commits infidelity. This danger is also present in the case of self-exclusion. A gambler might self-exclude, for example, because of the admonitions or threats of divorce from their spouse, a first-order decision that jars with their second-order preference to be more independent, to live a libertine life of indulgence, or simply to be a gambler. This heteronomous decision would bind the agent’s later autonomous decision to gamble. As deplorable as this person’s skewed priorities are, it is nonetheless the case that these priorities are their own. The point here is not that this is a regular occurrence but that, on the rare occasion it does occur, those enlisted to realise the Ulysses pact cannot be expected to discern the complex motivations which are obscured by the ostensibly simple self-exclusion agreement. In these instances, self-exclusion would become a vessel for the preservation of heteronomy, rather than a vessel for the realisation of autonomy.

The hierarchical approach cannot, I have argued, resolve the self-exclusion problem and, as such, I will now turn to an alternative approach to determining identification with a decision within a procedural framework of autonomy: the historical approach. Christman, an important proponent of a historical approach, contends that autonomy relies on “what the agent thinks about the process of coming to have the desire, and whether she resists that process when (or if) given the chance.” ^{Footnote 90} On this conception, the autonomy of a person’s decision is not contingent on some other, possibly heteronomous second-order preference. Rather, it relies on a historical construction of the self, in which one has developed in a procedurally independent way. Christman later adds the condition of non-alienation. To be alienated is to “experience strong negative affect relative to that characteristic—to disapprove in some manner—and to resist whatever motivational force it may have.” ^{Footnote 91} In essence, the agent is non-autonomous if aspects of their character would produce feelings of alienation and be rejected by the agent who critically reflected on how they came about. While it is not worthwhile to engage with all the minutiae of Christman’s theory here, ^{Footnote 92} I will now argue that it is beset by issues which make it equally as unable to solve the self-exclusion problem as the hierarchical approaches discussed earlier. One forceful challenge to Christman’s theory comes from Oshana, who questions whether alienation, as described by Christman, represents a failure of autonomy. ^{Footnote 93} She outlines the case of a reluctant careerist, who leaves their child with a childminder while they pursue full-time employment. This desire, Oshana illustrates, arises from an envious motivation to emulate the luxurious lives and conspicuous consumption of others. The reluctant careerist does not revise or reject that lifestyle but does abhor its origin in a selfish desire. Oshana argues that Christman must consider this agent non-autonomous because of the sense of alienation which arises from how their lifestyle began. She refutes the charge of heteronomy, however. The reluctant careerist may have made a choice with which they are uncomfortable, but they did not suffer from any cognitive failure in doing so. Oshana perhaps mischaracterises Christman’s point, however. Christman’s threshold for non-alienation is quite high. He contends that the agent must feel “no affinity” with the relevant aspect of themselves. ^{Footnote 94} The reluctant careerist, however, plainly has some affinity with their lifestyle; after all, they continue in it. On an all-things-considered judgement, the agent might feel distaste for some element of their motivation but ultimately has concluded that they identify with their choice.

A more forceful challenge to Christman’s account lies in alienating aspects which cannot be undone. In extremis, we might look to natural limitations. As Nozick writes, “I may voluntarily walk to someplace I would prefer to fly to unaided.” ^{Footnote 95} If I feel alienated by the inability to fly that comes with my birth as an ordinary human being, am I heteronomous? Plainly, this is untenable. Leaving aside this fantastic example, we might take the victim of a road traffic accident who has had a leg amputated as a result. The person might feel alienated by their accident, which certainly irretrievably limits their scope of choice in many or most aspects of their life. Must it be concluded that this person, on Christman’s understanding, has permanently and irreversibly been rendered heteronomous? Physical disability obviously poses unique challenges for theories of autonomy and therefore may be inappropriate grounds for objection to the historical approach. Given that the focus of this article is Gambling Disorder, what of psychiatric disorders? Take the example of a person suffering from Avoidant Personality Disorder (AVPD) as a result of non-ideal childhood experiences. ^{Footnote 96} The sufferer was too young to express any actual resistance to the traumatic experience which led to their AVPD and, as such, heteronomy cannot be ascertained on these grounds. Like the amputee, the sufferer’s traumatic experience cannot be undone, though its effects perhaps can. However, even if the person undergoes treatment to address their AVPD, their subsequent choices are those of a person who has undergone an autonomy-distorting event and psychiatric treatment to address it, not those of a person who has not undergone an autonomy-distorting event. Their choices can never be fully extricated from their traumatic experience. Does a historical approach therefore demand that a person who undergoes an unresisted, traumatic experience be forever categorised as heteronomous?

Christman, to his credit, addresses this concern. He writes that an assessment may need to be hypothetical, in the case of an agent who did not actually resist the development of a desire as it happened but “would have done so under conditions that make this possible.” ^{Footnote 97} Here, however, Christman veers inevitably into the substantive. Procedural autonomy is distinguished by its indifference to the content of decisions. If the autonomy or heteronomy of a decision is to be determined not by an agent’s actual circumstances and characteristics but by a hypothetical judgement, this can surely only be determined by whether a hypothetical agent would make a decision with the specific content at hand. This is an inevitably partial judgement and does not emanate, as procedural autonomy demands, from the actual agent. Whatever of the substantivity of Christman’s theory, its applicability to the self-exclusion problem is precluded by the same issue as hierarchical theories: its impracticability. For a historical approach to justify self-exclusion, it would have to hold the t ¹ decision to self-exclude as the historical, non-alienated and therefore autonomous preference and the t ² decision to gamble as alienated and therefore heteronomous. Of course, a gambling operator and its staff lack the relevant information to make this determination. In the case of a hypothetical judgement, they are even less able to determine what the self-excluder before them would have done. (In fact, no one is so able.)

In sum, neither a hierarchical nor a historical approach to preferences can resolve the self-exclusion problem. As a matter of abstract philosophy, either approach might theoretically justify self-exclusion by considering the t ¹ decision autonomous and the t ² decision heteronomous. In practice, however, the opacity of the self-excluder’s thought processes makes it impossible for operators and their staff—as enlistees to enforce the Ulysses pact—to reliably make this judgement. It must then be assumed, for practical purposes, that both the t ¹ and t ² decisions are equally autonomous.

IV. Proposed solutions to the self-exclusion problem

Thus far, I have argued that the self-exclusion problem cannot be resolved by considering the t ² decision heteronomous because of a failure of rationality or, more generally, capacity. This approach would deprive the t ¹ decision of any gravitas, which would fail to justify self-exclusion as a distinct, autonomy-based alternative to paternalistic, forced exclusion. Hierarchical and historical approaches to autonomy purport to solve this issue by providing an avenue by which the t ² decision can be categorised as heteronomous relative to the agent’s own preferences. At the theoretical level, these approaches’ merit is contestable. As regards the self-exclusion problem, their impracticability makes them irrelevant. As such, the self-exclusion problem remains, and we are left with two conflicting preferences of presumptively equal autonomy. I will now consider two attempts to resolve the issue of conflicting, autonomous preferences.

Davis succinctly explains the issue with what I have called the self-exclusion problem, although he does not refer to it as such or write in the context of problem gambling. ^{Footnote 98} He explains that self-binding in this way conflicts with what he calls the “Current Preference Thesis”, i.e., that respecting an individual’s autonomy means only respecting their current preferences. ^{Footnote 99} Accepting the Current Preference Thesis at face value would make self-exclusion impossible. At t ² , the pathological gambler’s preference would be to gamble, this ought to be respected and the self-exclusion agreement revoked or ignored. Davis’ solution to the existence of conflicting preferences—at t ¹ and t ² —is to identify a third preference, which he calls the “resolution preference” and which is a preference expressed by the individual on how to resolve a foreseen conflict of preferences. ^{Footnote 100} This initially appears appealing in resolving the self-exclusion problem. The gambler has two conflicting preferences at different points in time: at t ¹ , their preference is not to gamble once t ² arrives and, at t ² , their preference is to gamble at t ² . Their resolution preference is the self-exclusion agreement, which was presumably entered into in foresight of a future desire to gamble and which states that, in that case, the gambler ought to be refused service. Davis’ apparent solution evokes the hierarchical approaches discussed earlier. Rather than trying to determine which of the t ¹ and t ² decisions is the higher-order preference, however, both are cast as first-order preferences. These, then, are both subordinate to the second-order, resolution preference. The resolution preference demands that, in the case of conflict, the t ¹ preference ought to prevail. On the one hand, Davis’ approach avoids the impracticability of hierarchical approaches. Those enlisted to enforce the Ulysses pact need not identify which preference belongs to a higher order. Rather, the decisive resolution preference is explicitly set out by the self-exclusion agreement. On the other hand, it does not avoid the problem of infinite regress. A resolution preference is no more unimpugnable than a second-order preference in a hierarchical approach and an agent may therefore develop conflicting resolution preferences. Imagine a t ² gambler who is told that their resolution preference demands a refusal of service because their t ¹ preference to self-exclude must prevail. There is nothing to stop the gambler from simply saying that they now have a new resolution preference in favour of t ² . These conflicting resolution preferences would now necessitate recourse to an even higher-order resolution preference about conflicting resolution preferences, which could itself be confronted by a conflicting preference, and so on. Davis says that, in the case of an agent with conflicting resolution preferences, “it is impossible to respect his autonomy—but that is his problem, not ours.” ^{Footnote 101} This flippant dismissal, however, raises the question of why facilitating a problem gambler avoiding their vice is ‘our problem’ at all and why we have bothered to come even this far.

Another potential solution is advanced by Levy, who writes specifically in the context of addiction. ^{Footnote 102} He dismisses synchronic approaches—Frankfurt’s in particular—and adopts a diachronic view of autonomy as the agent’s extended will across time. ^{Footnote 103} According to Levy, humans typically unify themselves as they mature. He equates this with altering one’s discount curve for future goods until it approaches the ideal, so that one begins to value future goods appropriately in relation to immediate interests and maintain a unified will accordingly. ^{Footnote 104} Addicts, he says, are a partial exception to this; they are more susceptible to rapidly changing preferences (the very concept of self-exclusion exemplifies this), they are unable to apply their extended will over time and are, supposedly, more ‘fragmented’ than non-addicts. He says,

Levy separates one’s (extended) will from temporary preferences. ^{Footnote 106} Insofar, then, as he bases autonomy on historical personal development, Levy’s theory is susceptible to the same charges of impracticability I earlier levied against Christman’s. A gambler who enters into a self-exclusion agreement undoubtedly will have a history of gambling, one which may even outweigh in duration their period of desiring abstinence. How, then, are an operator and its staff to determine which of the t ¹ or t ² preferences is temporary and which reflects the agent’s will over time? If the will is simply identified with abstinence because it is preferable, this is plainly a substantive claim. Levy identifies this challenge and, drawing on Bratman, ^{Footnote 107} proposes to “identify the agent herself with the part-self which is the product of her own strong-arm tactics.” ^{Footnote 108} By this, he means that the individual’s ‘true self’ is the one that is forward-planning, and that this true self’s desire to abstain from their vice ought to be regarded as their extended will. This is nonetheless unsatisfying as a procedural theory of autonomy. Identifying the true self with forward planning and distinguishing this true self’s extended will from temporary preferences would probably justify self-exclusion. However, Bratman, from whose theory Levy admits to drawing heavily, holds that the true self’s acceptance of a desire is vindicated when the individual has a self-governing policy which views that desire as providing a justifying reason according to practical reasoning. From here, we can easily move to Finnis’ assertion that having a coherent plan of life forms an essential aspect of practical reasoning and therefore a good life. ^{Footnote 109} My intention is not to disagree with the importance Levy, Bratman, and Finnis place on planning, coherence, and self-governance. However, these are all plainly substantive assertions. In the case of a self-excluded gambler, the assumption that long-term planning is necessarily represented by the addict’s self-denial rather than, say, a lifelong commitment to libertinism is a substantive condition. This substantive condition gives preference to one value, self-denial, over another, indulgence, based on reasons which do not emanate from the agent and therefore cannot be considered a procedural approach. ^{Footnote 110}

V. Why self-exclusion is permissible and voluntary slavery is not

Thus far, I have demonstrated the failure of a procedural theory of autonomy to justify self-exclusion. A content-neutral, procedural theory demands that autonomous preferences emanate from the agent and are not based on the substantive preferences of others. As I have argued, procedural attempts to resolve the self-exclusion problem invariably fail either at the theoretical level—because they introduce substantive conditions or are beset by the problem of infinite regress—or at the practical level—because of their impracticability. I will now turn to the final point of my argument, which is to make explicit what has so far been implicit: self-exclusion can only be justified within a normative framework that considers abstinence from gambling as preferable to participation in gambling. To this end, I will compare self-exclusion to a topic which has troubled procedural theorists of autonomy for centuries, the “riddle” of voluntary slavery. ^{Footnote 111}

By voluntary slavery, I simply mean a contract in which one party agrees to become the slave of another in exchange for consideration of some sort. Voluntary slavery contracts bear some resemblance to self-exclusion. The person decides at t ¹ to become the slave of another person or to have themselves excluded from gambling, respectively. At t ² , the person articulates a desire contrary to their t ¹ decision—to be manumitted or to gamble—and is kept from fulfilling this desire because of the prior decision. ^{Footnote 112} As a matter of procedural autonomy, voluntary slavery contracts are presumably permissible as long as the prospective slave is not coerced or manipulated. Nonetheless, I assume that, while self-exclusion grows in popularity, voluntary slavery is widely considered abhorrent and unacceptable. How can this different treatment be justified, however? For Mill, voluntary slavery is unacceptable because of its totality:

The slave, he says, foregoes any future exercise of freedom. This, of course, is distinguishable from self-exclusion, which precludes only a narrow range of activities. Is it true, however, that slavery represents such a uniquely total relinquishment of future autonomy that it is to be distinguished from all else? Block points out that voluntary slavery agreements do not, in fact, involve the total transfer of autonomy. ^{Footnote 114} The voluntary slave transfers control over their body, giving their owner the right to enact violence on them if they seek to defy the owner’s commands and thereby foregoing bodily autonomy. The slave has not, however, transferred their will. The master can force the slave to toil in a mine for eighteen hours per day, seven days per week. The master can even compel the slave to act as though they are happy with this state of affairs. The master cannot, however, compel the slave to actually be happy. The will is, as a matter of fact, inalienable. ^{Footnote 115} Acknowledging that voluntary slavery does not therefore involve a uniquely total transfer of autonomy, its difference from self-exclusion appears to be merely a matter of degree.

If voluntary slavery is distinguishable from self-exclusion only by the extent to which it limits autonomy and not some a priori difference, how can their different treatment be justified? One approach might be to consider autonomy as an abstract principle to be maximised. This abstract principle of autonomy, seeking maximisation, will be privileged over any individual act of autonomy. Therefore, the potentially autonomous act of agreeing to a contract for voluntary slavery is void ab initio because it extinguishes the potential for a greater realisation of autonomy (perhaps measured in number of autonomous acts performed) in the future. This view certainly coheres with Mill’s utilitarian conception of autonomy, which differs markedly from deontological views of autonomy as a right. Rather, Mill defends autonomy for its instrumental, rather than intrinsic, value. He argues that vindicating autonomy contributes to utility by diffusing the potential sources of improvement, writing, “the only unfailing and permanent source of improvement is liberty, since by it there are as many possible independent centres of improvement as there are individuals.” ^{Footnote 116} This he calls the “progressive principle.”Footnote ¹¹⁷ The autonomy maximisation model, then, might strike down voluntary slavery contracts on the grounds that the autonomy realised in their creation is eclipsed by the future autonomy they extinguish.

Nonetheless, this model contains troubling implications. Firstly, it would compel individuals to engage in the paralysing enterprise of having to calculate in advance whether each of their actions would lead to maximal autonomy. Faced with any choice, the individual would have to predict which option would create the most opportunities for autonomous choice, then which choice presented by each of those options would do the same, and so on ad infinitum. This issue extends to a meta-level, too. The time it would take to make these calculations carries significant opportunity costs. A person would have to forego opportunities for autonomous action in favour of taking time to calculate maximal utility. These lost opportunities would also have to be factored into calculations, which would increase the time dedicated to calculation, resulting in a never-ending cycle. Secondly and relatedly, the autonomy maximisation model suffers from humans’ inherently limited knowledge. If I knew I would die tomorrow, it is entirely probable that entering into a voluntary slavery contract today would be the most utility-maximising decision. I might, for example, leave the proceeds of the contract to my children, ensuring their long-term financial security. It must be remembered that prohibiting a voluntary slavery contract also limits my autonomy insofar as I cannot enter into one. If I died tomorrow, limiting my autonomy to sign a voluntary slavery contract today might represent an overall detriment to the utility of the final two days of my life. The same is true of self-exclusion. A problem gambler determined to overcome their addiction might derive no utility from gambling today and significant utility from signing a self-exclusion agreement. Prohibiting self-exclusion would therefore have a net negative effect on that gambler’s utility today. If the gambler dies at midnight, without having yet suffered even a momentary relapse in their addiction, signing the agreement is certainly the utility-maximising choice. Of course, most people will not die tomorrow. Perhaps the utility maximisation justification of Mill’s opposition to voluntary slavery is feasible on a societal level. Some, in very unique circumstances, may lose utility to the prohibition on voluntary slavery. Society in general might, however, yield greater utility by having more people innovating through autonomous action and thereby maximising societal utility. Mill does seem to consider this societal aspect. He claims that competition between the progressive principle and custom is the “chief interest of the history of mankind.” ^{Footnote 118} Calculating utility at the societal level does not escape the challenge of limited knowledge, however. A voluntary slave’s master might put them to work on a project of such utility that it outweighs the utility lost through legalised voluntary slavery. Imagine a skilled medical researcher whose talent surpasses every other in their generation. The researcher, however, chooses to pursue a career as a jazz musician, despite their middling musical talent. Falling into significant debt in their chosen pursuit, a wealthy philanthropist induces the researcher-musician to sign a voluntary slavery contract and immediately puts the researcher to work developing a cure to a devastating and widespread plague. The utility of this cure would surely be greater to society than that lost through prohibiting voluntary slavery, not least in that it would preserve the future autonomy of those would otherwise have died.

It is worth restating that procedural theories of autonomy demand content-neutrality. A procedural approach therefore cannot object to voluntary slavery as impermissible because it is, for example, repugnant to human dignity. Nor, as I have argued, can a procedural approach refute voluntary slavery based on what I have called a utility maximisation model. The only remaining avenue, then, is a consistent preference for later decisions over prior ones. This approach, which would preclude Ulysses pacts entirely, is something Hobbes considered to be an essential aspect of sovereignty and therefore, presumably, autonomous self-sovereignty. ^{Footnote 119} There is some ostensible merit to this view. Notwithstanding akrasia, the later actor does have access to information which might have been unforeseen to their earlier self. A voluntary slave might rue their naïve earlier decision when confronted with the grim reality of slavery. Similarly, a self-excluded gambler may not have foreseen just how intense their craving would be or the embarrassment they feel upon being refused service in front of members of their community. A Hobbesian approach to self-sovereignty certainly provides grounds for prohibiting voluntary slavery. However, it also makes Ulysses contracts of any sort impossible; Ulysses, Uncle Bill, and the self-excluded gambler must all have their earlier commitment disregarded and be left to choose the Sirens’ enticements, cocaine, and gambling, respectively. Alternatively, of course, the t ¹ decision might instead be consistently preferred. This would allow for Ulysses pacts, including self-exclusion. However, within a content-neutral, procedural framework, it would also allow for voluntary slavery contracts. Robert Nozick, one of the few theorists to take this content-neutrality to its logical extreme, allows for voluntary slavery contracts, writing,

This typifies the issue with a content-neutral approach that allows for precommitment; as much as it permits precommitment of one sort, it must also allow precommitment of the opposite sort. A person can sell themselves into slavery or commit never to do so. Likewise, a problem gambler can commit to self-exclusion but can also commit never to self-exclude, a commitment which the gambler may be induced to make in exchange for special offers or which might simply become a standard clause in gambling contracts. This brings me to the main point of this article. I contend that self-exclusion is justifiable for the same reason that voluntary slavery is not—a normative position. In order for self-exclusion to be theoretically justifiable, it cannot be based on a content-neutral, procedural theory of autonomy. Rather, it must be based on an affirmative position—commonly acknowledged from the outset by all parties—that abstinence from gambling is preferable to indulgence.